In this meetup, we will approach advanced topics in detail, with a different approach to deep learning. It will be based on the paper “Parametric machines: a fresh approach to architecture search” (https://arxiv.org/abs/2007.02777).
Using tools from category theory, we provide a framework where artificial neural networks, and their architectures, can be formally described. We first define the notion of machine in a general categorical context and show how simple machines can be combined into more complex ones. We explore finite- and infinite-depth machines, which generalize neural networks and neural ordinary differential equations. Borrowing ideas from functional analysis and kernel methods, we build complete, normed, infinite-dimensional spaces of machines, and discuss how to find optimal architectures and parameters–within those spaces–to solve a given computational problem. In our numerical experiments, these kernel-inspired networks can outperform classical neural networks when the training dataset is small.
We have the pleasure of announcing Mattia Bergomi – Chief Scientific Officer and Principal Investigator at Veos Digital, and Pietro Vertechi AI researcher at Veos Digital, – as this meetup’s speakers (https://www.linkedin.com/in/mattia-bergomi-54702b62/).
As always, expect up to 1h of talk, with around 10 minutes in the end for the Q&A (be sure to leave all the questions you might have during the talk on the YouTube chat!).
The presentation content will be available on our GitHub page (https://github.com/DeepLearningLisbon). The video will be streamed and saved on our Youtube page (https://www.youtube.com/channel/UCo_N8WcUxAdGFB8CCQcGr2A). You’ll also be able to see this meetup’s streaming link after registering for the event.
We hope that you’ll join us!